Backend interview prep

Start

Cassandra Architecture Interview Questions

Master Apache Cassandra architecture with interview questions covering Cluster, Node, Ring Architecture, Gossip Protocol, VNodes, Commit Log, Memtable, SSTable, Read/Write Path, Compaction, Repair, and enterprise production concepts.

Introduction

Apache Cassandra is one of the most advanced distributed databases ever built. Unlike traditional relational databases that depend on a master server, Cassandra follows a masterless peer-to-peer architecture, allowing every node to participate equally.

Its architecture enables

  • Linear Scalability
  • High Availability
  • Fault Tolerance
  • Multi-Data Center Replication
  • Zero Single Point of Failure

Understanding Cassandra architecture is essential for senior backend engineers, database engineers, solution architects, and distributed systems interviews.

Cassandra Cluster Architecture

flowchart LR

Client --> CoordinatorNode

CoordinatorNode --> Node1

CoordinatorNode --> Node2

CoordinatorNode --> Node3

Node1

<-->Node2

Node2

<-->Node3

Node3

<-->Node1

1. What is Cassandra Architecture?

Answer

Cassandra follows a

  • Distributed Architecture
  • Peer-to-Peer Architecture
  • Masterless Architecture

Every node performs

  • Reads
  • Writes
  • Replication
  • Communication

There is no master database server.

2. What is a Cassandra Cluster?

A Cluster is a collection of Cassandra nodes working together.

Example

Cluster

├── Node1

├── Node2

├── Node3

└── Node4

The cluster behaves like one logical database.

3. What is a Node?

A Node is a single Cassandra server.

Responsibilities

  • Store Data
  • Process Reads
  • Process Writes
  • Replicate Data
  • Participate in Gossip

4. What is a Data Center?

A Data Center is a logical group of nodes.

Example

US-East

US-West

Europe

Asia

Each data center can have its own replication factor.

5. What is a Rack?

A Rack is a group of nodes inside a data center.

Example

Data Center

├── Rack 1

│     ├── Node1

│     └── Node2

├── Rack 2

│     ├── Node3

│     └── Node4

Rack awareness improves fault tolerance.

6. Why does Cassandra use Peer-to-Peer Architecture?

Benefits

  • No Single Point of Failure
  • High Availability
  • Easy Horizontal Scaling
  • Automatic Failover

7. What is a Coordinator Node?

The Coordinator Node receives client requests.

Responsibilities

  • Route Read Requests
  • Route Write Requests
  • Collect Responses
  • Return Final Result

Any node can become a coordinator.

Coordinator Flow

flowchart LR

Client --> Coordinator --> Replica1

Coordinator --> Replica2

Coordinator --> Replica3

8. What is the Cassandra Ring?

All nodes logically form a ring.

Node1

↓

Node2

↓

Node3

↓

Node4

↓

Node1

The ring determines data distribution.

9. What is Token?

Each node owns one or more token ranges.

Example

0

100

200

300

Partition Keys are hashed into tokens.

10. What is a Partitioner?

Partitioner converts

Partition Key

↓

Token

Common Partitioner

  • Murmur3Partitioner

Token Distribution

flowchart LR

PartitionKey --> HashFunction --> Token --> ResponsibleNode

11. What are Virtual Nodes (vnodes)?

Instead of one token,

Each node owns multiple token ranges.

Benefits

  • Better Load Balancing
  • Faster Node Recovery
  • Easier Cluster Expansion

VNode Example

Node1

Token 10

Token 200

Token 800

Node2

Token 50

Token 300

Token 900

12. What is Gossip Protocol?

Gossip is Cassandra's internal communication protocol.

Purpose

  • Node Discovery
  • Health Monitoring
  • Cluster Membership
  • Failure Detection

Runs every second.

Gossip Architecture

flowchart LR

Node1

<-->Node2

Node2

<-->Node3

Node3

<-->Node4

Node4

<-->Node1

13. What are Seed Nodes?

Seed Nodes help new nodes discover the cluster.

Important

Seed nodes are

NOT

Master nodes.

14. What is Snitch?

Snitch provides topology information.

It understands

  • Rack
  • Data Center

Popular Snitches

  • GossipingPropertyFileSnitch
  • SimpleSnitch

15. What is Commit Log?

Every write is first stored in Commit Log.

Purpose

Crash Recovery

Write Path

flowchart LR

Client --> CommitLog --> Memtable --> SSTable

16. What is Memtable?

Memtable is an in-memory write buffer.

Once full

↓

Flushed to SSTable

17. What is SSTable?

SSTable

Sorted String Table

Immutable disk storage file.

Benefits

  • Fast Reads
  • Sequential Writes
  • No In-place Updates

Storage Architecture

flowchart LR

CommitLog --> Memtable --> SSTable

18. What is Bloom Filter?

Bloom Filter quickly determines

Whether data

may exist

inside an SSTable.

Benefits

  • Reduce Disk Reads
  • Faster Queries

19. What is Partition Index?

Helps Cassandra locate partition data quickly.

20. What is Index Summary?

Stores sampled partition indexes in memory.

Reduces memory usage.

21. What is Compression?

Compresses SSTables on disk.

Benefits

  • Save Storage
  • Faster Disk Reads

22. Explain Cassandra Write Path.

flowchart LR

Client --> Coordinator --> CommitLog --> Memtable --> SSTable

Steps

  1. Coordinator receives request.
  2. Commit Log stores data.
  3. Memtable stores data in memory.
  4. Data flushes to SSTable.

23. Explain Cassandra Read Path.

flowchart LR

Client --> Coordinator --> BloomFilter --> Memtable --> SSTables --> Result

Steps

  • Check Memtable
  • Check Bloom Filter
  • Read SSTables
  • Merge Results

24. Why are SSTables Immutable?

Benefits

  • No Locks
  • Fast Writes
  • Sequential IO
  • Better Performance

25. What is Compaction?

Compaction merges multiple SSTables.

Benefits

  • Remove Tombstones
  • Merge Updates
  • Improve Read Performance

Compaction Flow

flowchart LR

SSTable1 --> Compaction

SSTable2 --> Compaction

SSTable3 --> Compaction

Compaction --> NewSSTable

26. Types of Compaction

  • SizeTieredCompactionStrategy (STCS)
  • LeveledCompactionStrategy (LCS)
  • TimeWindowCompactionStrategy (TWCS)

27. What is Read Repair?

If replicas contain inconsistent data

↓

Coordinator repairs them during reads.

28. What is Hint Handoff?

If one replica is unavailable,

Coordinator stores

Hint

Later delivers data.

Hint Handoff Flow

flowchart LR

Write --> ReplicaDown --> HintStored --> ReplicaOnline --> ReplayHint

29. What is Anti-Entropy Repair?

Synchronizes replicas.

Uses

Merkle Trees

Command

nodetool repair

30. What are Merkle Trees?

Tree structure used to compare replica data efficiently.

Only differences are synchronized.

31. What is Failure Detection?

Uses Gossip protocol.

Nodes exchange heartbeat information.

Unresponsive nodes become

DOWN.

32. What happens when a node fails?

  • Coordinator reroutes requests.
  • Replicas continue serving data.
  • Hint Handoff stores missed writes.
  • Repair synchronizes data later.

33. What is Bootstrap?

Adding a new node to the cluster.

The node automatically receives token ranges.

34. What is Decommission?

Gracefully removes a node.

Data is redistributed.

35. What is Rebuild?

Copies data from another data center.

Useful for

Multi-DC deployments.

Enterprise Architecture

flowchart LR

Application --> Coordinator --> Cluster

Cluster --> CommitLog

Cluster --> Memtable

Cluster --> SSTables

Cluster --> Repair

Cluster --> Replication

Enterprise Best Practices

  • Use multiple racks.
  • Use NetworkTopologyStrategy.
  • Keep seed nodes stable.
  • Monitor Gossip health.
  • Use virtual nodes.
  • Schedule regular repairs.
  • Choose proper compaction strategy.
  • Monitor disk usage.
  • Monitor SSTable count.
  • Avoid oversized partitions.

Quick Revision

Component Purpose
Cluster Collection of Nodes
Node Database Server
Coordinator Handles Client Requests
Ring Logical Data Distribution
Token Data Ownership
Partitioner Generates Tokens
Gossip Node Communication
Seed Node Cluster Discovery
Snitch Topology Awareness
Commit Log Crash Recovery
Memtable In-Memory Buffer
SSTable Immutable Disk Storage
Bloom Filter Avoid Disk Reads
Compaction Merge SSTables
Hint Handoff Store Missed Writes
Read Repair Fix Replica Differences
Repair Synchronize Data

Interview Tips

Interviewers commonly ask

  • Explain Cassandra Write Path.
  • Explain Cassandra Read Path.
  • What happens when a node fails?
  • Why is Cassandra masterless?
  • Explain Gossip Protocol.
  • Explain Commit Log.
  • Why are SSTables immutable?
  • Explain Compaction strategies.
  • Explain Hint Handoff.
  • Explain Read Repair vs Repair.
  • Explain Virtual Nodes.
  • Why are Seed Nodes required?

Drawing the write path and read path diagrams during interviews often demonstrates a strong understanding of Cassandra internals.

Summary

Apache Cassandra's architecture is built around a peer-to-peer, masterless distributed design that provides high availability, fault tolerance, and linear scalability. Components such as Gossip Protocol, Commit Log, Memtables, SSTables, Bloom Filters, Compaction, Read Repair, Hint Handoff, and Virtual Nodes work together to deliver reliable performance for large-scale distributed applications.

Understanding these architectural components is essential before exploring advanced topics such as Data Modeling, Consistency Levels, Replication Strategies, and Performance Tuning in the upcoming chapters.